17/08/2023
Erik Kusch
Erik Kusch
Senior Engineer
Machine Readable Nature Research Group (MANA)
Department of Research and Collections
Natural History Museum
University of Oslo
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
1
Enhancing Projections of Biodiversity in the Anthropocene
ECOLOGICAL NETWORKS IN
MACROECOLOGICAL
RESEARCH
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
2
MOTIVATION & BACKGROUND
Biological Interactions and Macroecology in the Anthropocene
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
3
Biodiversity Is Connected and Under Threat
Illustration credit: http://www.davidebonadonna.it/
?
?
?
?
?
?
?
Extinctions ripple through the
biosphere via ecological interactions.
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
3
Biodiversity Is Connected and Under Threat
Illustration credit: http://www.davidebonadonna.it/
?
?
?
?
?
?
?
Extinctions ripple through the
biosphere via ecological interactions.
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
3
Biodiversity Is Connected and Under Threat
Extinctions ripple through the
biosphere via ecological interactions.
?
?
?
?
?
?
?
≊
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
3
Biodiversity Is Connected and Under Threat
Extinctions ripple through the
biosphere via ecological interactions.
?
?
?
?
?
?
?
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
3
Biodiversity Is Connected and Under Threat
Extinctions ripple through the
biosphere via ecological interactions.
?
?
?
?
?
?
?
Anthropocene & Ecosphere
▪
Environmental conditions
▪
Expressions of species interactions
▪
Species extinctions
▪
...
The Ecosphere is being reshuffled
▪
Local
▪
Regional
▪
Continental
▪
Global
Changes
Scales
▪
Macro-scale threat to ecological communities = threat to humans
Humanity is dependant on stability of
ecosystems & ecological communities
Effects and threats
across geographic scales
Macroecological perspective relevant for study-needs of the Anthropocene.
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
4
Central Research Question
Under the Anthropocene, how can we use ecological interactions to understand
ecosystem stability and forecast ecosystem change at macroecological scales?
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
5
An ecological interaction is the effect of one species on...
▪
Ecological interactions are highly complex:
▪
Sign (+/-)
▪
Magnitude
▪
Directionality (directed vs. undirected)
▪
Ecological interactions can be abstracted:
▪
Networks/Graphs:
▪
Nodes/Vertices = Species Identities
▪
Links/Edges = Interactions
▪
Network Matrices:
▪
Columns/Rows = Species Indetities
▪
Cell values = Interactions
Ecological Interactions – Complexity & Abstraction
... the probability of occurrence of another species.
... the fitness of another co-occurring species.
A
B
C
D
E
A
-1
3
B
1
C
-3
D
-2
E
A)
B)
A
B
C
D
E
A
-1
B
C
3
D
E
1
-3
-2
A)
B)
Node
Link
Node
Ecological interactions can be studied via
ecological networks.
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
6
Central Research Needs – Limiting Scope
Under the Anthropocene, how can we use ecological interactions to understand
ecosystem stability and forecast ecosystem change at macroecological scales?
Under the Anthropocene, how can we use ecological networks to understand
ecosystem stability and forecast ecosystem change at macroecological scales?
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
6
Central Research Needs – Limiting Scope
Under the Anthropocene, how can we use ecological networks to understand
ecosystem stability and forecast ecosystem change at macroecological scales?
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
6
Central Research Needs – Limiting Scope
Under the Anthropocene, how can we use ecological networks to understand
ecosystem stability and forecast ecosystem change at macroecological scales?
▪
Explore likely extinction cascade
scenarios
▪
Evaluate current macroecological
research and data practices
▪
Evaluate statistical methodology
for macroecological networks
Updating Macroecological
Rersearch Practices.
Chapter I
Using Ecological Networks as
Forecast Tools.
Chapter II
Inferring Biological Interactions
from Proxies.
Chapter III
Climate Change.
Extinction Cascades.
Macroecological Networks.
I. The Anthropocene II. Ecosystem Change III. Macroecologial Scales
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
7
Updating Macroecological Research Practices
Chapter I
Evaluating current macroecological research and data practices
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
8
The Holy Trinity of Abiotic Data in the 21
st
Century
Variables
Accuracy
Resolution
Essential
Climate
Variables
(ECVs)
Data
Accuracy
Spatial &
Temporal
Resolution
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
9
Accuracy:
Good fit due to data assimilation practices
Provisioning of uncertainty metrics
Reformation: Climate Reanalyses are the Solution
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
9
Accuracy:
Good fit due to data assimilation practices
Provisioning of uncertainty metrics
Resolution:
Space: 9km (ERA5-Land)
Time: hourly intervals
Reformation: Climate Reanalyses are the Solution
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
9
Accuracy:
Good fit due to data assimilation practices
Provisioning of uncertainty metrics
Resolution:
Space: 9km (ERA5-Land)
Time: hourly intervals
ECVs offered by climate reanalyses:
e.g.: ERA5(-Land): ~83 variables
Covering all ECVs indexing important components of
ecosystems such as:
Atmosphere
Soil properties
Reformation: Climate Reanalyses are the Solution
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
10
RESOLVING ROADBLOCKS TO USING
CLIMATE REANALYSIS DATA IN R
WITH KrigR
Published January 6, 2022
10
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
11
KrigR – Novel Climate Data Stream
Data Retrieval
Spatial Resolution
ERA5(-Land) data retrieval is too complex
and unintuitive for many users.
9x9km resolution is too coarse for some
downstream applications.
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
12
KrigR – Novel Climate Data Stream (Downloads)
Data Retrieval
Era5(-Land) Data /
Spatial product
▪
More intuitive download
specification
▪
Spatial data limiting
beyond extents
▪
Aggregation to desired
temporal resolutions
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
12
KrigR – Novel Climate Data Stream (Downloads)
Data Retrieval
Era5(-Land) Data /
Spatial product
Intuitive download specification
Spatial data limiting
Desired temporal resolutions & aggregate metrics
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
12
KrigR – Novel Climate Data Stream (Downloads)
Data Retrieval
Era5(-Land) Data /
Spatial product
Intuitive download specification
Spatial data limiting
Desired temporal resolutions & aggregate metrics
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
12
KrigR – Novel Climate Data Stream (Downloads)
Data Retrieval
Era5(-Land) Data /
Spatial product
Intuitive download specification
Spatial data limiting
Desired temporal resolutions & aggregate metrics
17/08/2023
Erik Kusch
download_ERA() can take shapefiles
and point-locations.
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
12
KrigR – Novel Climate Data Stream (Downloads)
Data Retrieval
Era5(-Land) Data /
Spatial product
Intuitive download specification
Spatial data limiting
Desired temporal resolutions & aggregate metrics
17/08/2023
Erik Kusch
download_ERA() can take shapefiles
and point-locations.
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
12
KrigR – Novel Climate Data Stream (Downloads)
Data Retrieval
Era5(-Land) Data /
Spatial product
Intuitive download specification
Spatial data limiting
Desired temporal resolutions & aggregate metrics
17/08/2023
Erik Kusch
download_ERA() can take shapefiles
and point-locations.
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
12
KrigR – Novel Climate Data Stream (Downloads)
Data Retrieval
Era5(-Land) Data /
Spatial product
Intuitive download specification
Spatial data limiting
Desired temporal resolutions & aggregate metrics
17/08/2023
Erik Kusch
download_ERA() can take shapefiles
and point-locations.
Using the arguments TResolution,
TStep , and FUN download_ERA()
can aggregate time-series to any
desired temporal resolution.
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
12
KrigR – Novel Climate Data Stream (Downloads)
Data Retrieval
Era5(-Land) Data /
Spatial product
Intuitive download specification
Spatial data limiting
Desired temporal resolutions & aggregate metrics
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
13
KrigR – Novel Climate Data Stream (Downscaling)
Data Retrieval
Era5(-Land) Data /
Spatial product
Spatial Resolution
9x9km resolution is too coarse for some
downstream applications.
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
13
KrigR – Novel Climate Data Stream (Downscaling)
Data Retrieval
Era5(-Land) Data /
Spatial product
Downscaled
Product
Downscaling
Standard Error
&
&
Covariates
(Target Resolution)
Covariates
(Training Resolution)
&
Kriging
Covariates
Spatial Resolution
Downscaled
Product
Downscaling
Standard Error
&
Kriging
&
Covariates
(Target Resolution)
Covariates
(Training Resolution)
&
Covariates
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
13
KrigR – Novel Climate Data Stream (Downscaling)
Spatial Resolution
Downscaled
Product
Downscaling
Standard Error
&
Kriging
&
Covariates
(Target Resolution)
Covariates
(Training Resolution)
&
Covariates
KrigR provides USGS GMTED 2010 digital
elevation model data as interpolation
covariates.
krigR() enables parallel processing of multi-
layer rasters and allows for pausing and
restarting kriging via temporary files.
Kriging uncertainty
can help understand
quality and
robustness of
interpolated data.
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
14
KrigR – Novel Climate Data Stream
Downscaled
Product
Downscaling
Standard Error
&
&
Covariates
(Target Resolution)
Covariates
(Training Resolution)
&
Kriging
Covariates
Spatial Resolution
Data Retrieval
Era5(-Land) Data /
Spatial product
Third-
party
data
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
15
Using KrigR for Macroecological Studies
Data Stream vs. Static Product
Benefit from any improvements to source data
Encourages interdisciplinary engagement
Customisation of Data Products
Unparalleled potential for data customisation
Temporal aggregation & metrics
ECV provisioning
Align data products with research questions
Uncertainty Indicators
First workflow fully reporting data uncertainty
Propagation into downstream analyses?
Establishing bias corrected projections with KrigR:
Projection products:
downscaled future – downscaled historical = anomalies
Reanalysis products:
anomalies + downscaled reanalysis = projected reanalysis
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
16
REAL-WORLD APPLICATIONS OF
THE KrigR-PIPIELINE
Published November 15, 2021
16
I
Does Kriging perform well at statistically
downscaling macroecologically relevant data?
II
Do KrigR products differ significantly from static
legacy data products?
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
17
Kriging Accuracy
Testing Kriging Accuracy:
1. Krig upscaled ERA5-Land data to native resolution
2. Difference of upscaled & interpolated product ( )
3. Total uncertainty of kriged product:
𝜎
𝐸𝑟𝑟
𝜎
𝑇𝑜𝑡𝑎𝑙
=
𝜎
2
𝐾𝑟𝑖𝑔
+ 𝜎
2
𝐷𝑦𝑛
2
3. Where Kriging is not the most accurate method,
it is the only one that produces uncertainty estimates.
1. Kriging outperforms most other interpolation methods.
2. Kriging is highly accurate for a variety of ECVs.
I
Does Kriging perform well at statistically
downscaling macroecologically relevant data?
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
18
KrigR & Legacy Products
4. KrigR-products do not align with most legacy products.
5. Particularly, in topographically heterogenous regions,
KrigR seems most reliable (i.e. accurate) and informative
(through provision of uncertainty metrics) to us.
II
Do KrigR products differ significantly from static
legacy data products?
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
19
Implications for macroecology
19
Instead, macroecology ought to undergo a paradigm shift towards data streams making use of
newly developed, state-of-the-art climate reanalysis data and novel methdology such as KrigR.
The practice of using static data sets reporting only a small number of ECVs at subpar temporal resolutions is not
suited for macroecological research needs of the Anthropocene.
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
20
Using Ecological Networks as Forecast Tools
Chapter II
Exploring likely extinction cascade scenarios
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
21
Extinction Cascades & Proxies of Threat
?
?
?
?
?
?
?
Global
▪
IUCN Red List assessments
▪
…
Proxies for Risk of Primary Extinction
Localised
▪
Climate Change & Projections:
▪
Safety Margins
▪
...
▪
Placement in Network:
▪
Centrality
▪
…
What about secondary extinction risk?
P1
P2
P3
A1 A2 A3
P4
Initial Network
Legend
Post-Extinction Network
P1
P2
P3
A1 A2 A3
P4
I
Is choice of primary extinction
risk proxy relevant for
extinction cascade outcomes?
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
22
▪
Link-Loss-Sensitivity
▪
% of combined initial
interaction magnitude a
species requires for
continued existence
▪
Rewiring Probability
Thresholds
▪
Probability threshold
which a potential
interaction needs to
exceed for establishment
Secondary Extinction Risk: Network Resilience
Link-loss sensitivity and rewiring probability thresholds establish
two-dimensional landscapes of network resilience.
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
22
▪
Link-Loss-Sensitivity
▪
% of combined initial
interaction magnitude a
species requires for
continued existence
▪
Rewiring Probability
Thresholds
▪
Probability threshold
which a potential
interaction needs to
exceed for establishment
Secondary Extinction Risk: Network Resilience
II
Does considering network
resilience metrics enhance
extinction cascade analyses?
Link-loss sensitivity and rewiring probability thresholds establish
two-dimensional landscapes of network resilience.
Contemporary R workflows fail to
incorporate link-loss sensitivity or
rewiring thresholds.
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
23
Published June 2, 2023
23
NETWORK RESILIENCE MECHANISMS IN
EXTINCTION CASCADE ANALYSES IN R
WITH NetworkExtinction
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
24
Package Scope and Functionality
NetworkExtinction supports:
▪
Degree Distribution
▪
Extinction Cascade Visualisations
▪
Extinction Cascade Simulations
▪
Static order:
▪
Ordered
▪
Random
▪
Dynamic order
▪
Mostconnected
▪
Leastconnected
Supported Network Types:
▪
Mutualistic
▪
Trophic
Fully integrates:
▪
Link-loss sensitvity
▪
Rewiring Probability Thresholds
Defined either:
▪
Whole-network
▪
Individual nodes
Full exploration of network resilience
landscapes now possible.
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
25
EXPLORING EMPIRICAL NETWORK
RESILIENCE LANDSCAPES AT
MACROECOLOGICAL EXTENTS
Registered as preprint and under prepration for submission
25
I
Is choice of primary extinction risk proxy
relevant for extinction cascade outcomes?
II
Does considering network resilience metrics
enhance extinction cascade analyses?
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
26
Study Design & Primary Extinction Risk Proxies
Network Data
Plant-Frugivory Network Collection (Fricke et al., 2022)
406 networks
CMIP6
Projections
KrigR
rgbif
GBIF
1982-1999 Climatologies
SSP245 & SSP 585 2081-2100
1982-1999 Presence Records
1724 species
Climate Preferences & Niches
𝐶
𝑖,𝑗
=
𝜇
𝑖
− 𝑃
𝑗
𝜎
𝑖
𝐶
𝑖,𝑗
> 2
Species
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
Climate Safety Margins
IUCN
Centrality
IUCN
𝐼 𝑈 𝐶𝑁
𝑖
> 𝑒 𝑛 𝑑𝑎 𝑛 𝑔 𝑒 𝑟 𝑒 𝑑
𝐶𝑒𝑛 𝑡 𝑟 𝑎 𝑙𝑖𝑡 𝑦
𝑖,𝑗
∈ 𝑄 𝑢 𝑎 𝑛 𝑡 𝑖 𝑙𝑒 (𝐶 𝑒 𝑛 𝑡 𝑟 𝑎 𝑙𝑖𝑡 𝑦
𝑖
)
ConR
rredlist
igraph
Protected Areas
Extinction Risk Proxies
Networks for Analysis
81 networks
Species Richness
𝑛
𝑝𝑙𝑎𝑛𝑡𝑠
> 7
𝑛
𝑎𝑛𝑖𝑚𝑎𝑙𝑠
> 7
Interaction Strength Variation
𝜎
𝜇
> 0.5
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
27
Potential Networks & Rewiring Probability
Functional Trait Data
Functional Trait Expressions (Fricke et al., 2022)
124,179 records, 4715 species
Functional Expressions for Target Species
60,047 records, 1724 species
Subset for 1724
target species
identities
Plant
Plant-Trait 1
Animal
Animal-Trait …
Link Presence
Network ID
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
Rewiring Potential Identification
randomForest
Probability Matrices of Interactions between Target Species
Network 1
Plant A
Plant …
Animal A
…
…
…
Animal B
…
…
…
Animal …
…
…
…
Network 2
Plant A
Plant …
Animal A
…
…
…
Animal B
…
…
…
Animal …
…
…
…
Network …
Plant A
Plant …
Animal A
…
…
…
Animal B
…
…
…
Animal …
…
…
…
NetworkExtinction
Climate Safety Margins
IUCN
Centrality
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
28
Primary Extinction Species Pools
▪
Global extinction risk proxies (i.e. IUCN) identify very few
species for primary extinction pools
▪
Localised extinction risk proxies (i.e., Climate Safety Margins &
Centrality) identify much larger species pools for primary
extinction
Considerable overlap between Centrality and Climate Safety
Margin classifications
Climate Safety Margins
IUCN
Centrality
1. Primary extinction risk proxies at too coarse a scale may
underestimate biodiversity loss at the level of individual
empirical ecological networks.
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
29
Drivers of Threat to Global Ecological Networks
Effect of Extinction Proxy
Effect of Extinction-Cascade Directionality
2. Relative species loss post-extinction cascade simulation is considerably
higher for localised primary extinction risk proxies.
3. Arguing over whether to focus on bottom-up or top-down
extinction cascades is obfuscating the much bigger risk
posed by bi-directional extinction cascades.
I
Is choice of primary extinction risk proxy
relevant for extinction cascade outcomes?
Loss of Species
Loss of Species
SSP245
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
30
Empirical Network Resilience Landscapes
▪
Network resilience landscapes demonstrate
variation of future ecosystem scenarios
Contemporary practices fail to explore these
▪
Contemporary analysis assumptions:
▪
Secondary extinction when all links are lost
▪
No rewiring
Rewiring Probability Threshold
Relative Loss of Species
Link-Loss Sensitivity
Contemporary Extinction
Cascade Analyses
4. Contemporary simulations of extinction cascades likely
underpredict future biodiversity loss.
II
Does considering network resilience metrics
enhance extinction cascade analyses?
SSP585
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
31
Implications for ecological network forecasting
31
Even at macroecological scale, local-scale metrics are highly informative.
To use ecological networks for forecasting of ecological community structures, research ought to explore
realistic ranges in the network resilience landscapes of link-loss sensitivity & rewiring probability thresholds.
Neglecting to account for link-loss sensitivity and rewiring probability thresholds leads to a
likely underprediction of future biodiversity loss thus overestimating ecosystem stability.
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
32
Inferring Biological Interactions from Proxies
Chapter III
Evaluating statistical methodology for macroecological networks
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
33
▪
Observing ecological interactions is labour-intensive
Intractable at macroecological scales
▪
Multiple frameworks for association/interaction
inference have been proposed
Most analyse co-occurrence patterns
▪
Co-Occurrence based methods for association/
interaction inference have been critiqued
▪
Research Questions:
Inference of Ecological Networks
Statistical inference of ecological networks
II
Are inference methods and
their networks scalable?
I
Does choice of inference method affect
inferred network structure?
III
How accurate is network inference?
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
34
Philosophical Differences in Network Inference
Biodiversity Input Data Type
Consideration of
Environmental Conditions
Co-Occurrence
Abundance
Performance
Explicit
Implicit
None
Additional considerations of network inference:
▪
Cosideration of Environmental Conditions:
▪
Environmental conditions have been demonstrated to affect
expressions of interactions in identity and magnitude
▪
Spectrum of Co-occurrence– Performance:
▪
Statistically useable information content changes drastically
when considering presence/absence, abundance, or performance
All ecological network inference is built on spatial or temporal biodiversity data.
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
35
CONTRASTING ECOLOGICAL
NETWORK INFERENCE ACROSS
SCALES AND APPROACHES
Registered as preprint and currently under peer-review
35
II
Are inference methods and
their networks scalable?
I
Does choice of inference method affect
inferred network structure?
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
36
Study Design & Data Streams
# 810,919 at 160401 plots
# 70,797 at 640 plots
Yosemite Forest Dynamics Plot
Yosemite National Park
Tem perate Con ifer
Forest Biome
Pre-fire event in 2013
Raw
Data
Subsetting in
Time and Space
# 34,444 at 640 plots
11 species
Plot-Scale
# 291 at 101 plots
13 species
Region-Scale
# 96,169 at 46,328 plots
15 species
Macro-Scale
BIEN
YFDP Traits
KrigR
V.PhyloMaker
Plot
Temperature
Soil
Moisture
Precipi-
tation
Evapo-
ration
…
…
…
…
…
…
…
…
…
…
Species
SLA
Leaf Carbon
Leaf Nitrogen
…
…
…
…
…
…
…
…
Plot
Species 1
Species 2
Species …
…
…
…
…
…
…
…
…
Phylogeny
Distributional
Null Expectation
Environmental
Conditions
Spatial Products
Functional Trait
Data
Species
Plot
Performance
Abundance
Presence/ Absence
…
…
…
…
…
…
…
…
…
…
Adding Data for Use in
Ecological Network Inference
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
37
Network Inference Methods
Consideration of
Environmental Conditions
Co-Occurrence
Abundance
Performance
Explicit
Implicit
None
COOCCUR
NETASSOC
HMSC
HMSC
HMSC
Biodiversity Input Data Type
NDD-RIM
COOCCUR
NETASSOC
HMSC
NDD-RIM
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
38
No Consensus of Inference Across Approaches
I
Does choice of inference
method affect
inferred network structure?
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
39
No Consensus Across Scales
II
Are inference methods and
their networks scalable?
NETASSOC
HMSC
COOCCUR
NDD-RIM
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
40
What Drives Inference Dissimilarity?
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
40
What Drives Inference Dissimilarity?
Consideration of environmental
conditions a driving factor of
inference outcome at macro scale?
Performance information may
stabilise inference outcome.
But how do we know which inference approach
yields the most accurate results?
III
How accurate is network inference?
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
41
EVALUATING NETWORK INFERENCE
PERFORMANCE USING SYNTHETIC
SPATIAL PRODUCTS
Registered as preprint and under prepration for submission
41
III
How accurate is network inference?
IV
What drives differences in
network inference accuracy?
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
42
Study Concept & Simulation Set-Up
▪
Need to know the “true” ecological network to
evaluate network inference accuracy
Random generation of ecological network
▪
Spatial biodiversity data analysed by the assessed
inference approach ought to reflect:
▪
Environmental conditions & species-specific niche preferences
▪
Interactions between individuals of interacting species
Demographic simulation with variable death rate:
𝑑
𝑖
(
𝑡
)
= 𝑑
0
+ 𝑃
𝑖
× 𝐸
𝑖
− Ω
𝑖
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
43
Variable Death Rate Components
𝑃
𝑆
(𝑡) =
𝑁
𝑆
(𝑡) × (𝑏
0
− 𝑑
0
)
𝐾
𝑆
𝑑
𝑖
(
𝑡
)
= 𝑑
0
+ 𝑃
𝑖
× 𝐸
𝑖
− Ω
𝑖
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
43
Variable Death Rate Components
𝑃
𝑆
(𝑡) =
𝑁
𝑆
(𝑡) × (𝑏
0
− 𝑑
0
)
𝐾
𝑆
𝐸
𝜇
(x, y) = 𝑀
𝜎
𝐸
(
𝜇
𝑖
− 𝑢
0
(
𝑥, 𝑦
)
)
𝑀
𝜎
𝐸
(
𝑧
)
= 𝑒
(
𝑧
𝜎
𝐸
)
2
𝑑
𝑖
(
𝑡
)
= 𝑑
0
+ 𝑃
𝑖
× 𝐸
𝑖
− Ω
𝑖
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
43
Variable Death Rate Components
𝑃
𝑆
(𝑡) =
𝑁
𝑆
(𝑡) × (𝑏
0
− 𝑑
0
)
𝐾
𝑆
𝐸
𝜇
(x, y) = 𝑀
𝜎
𝐸
(
𝜇
𝑖
− 𝑢
0
(
𝑥, 𝑦
)
)
𝑀
𝜎
𝐸
(
𝑧
)
= 𝑒
(
𝑧
𝜎
𝐸
)
2
Ω
𝑖
=
∑
𝐽
1
𝑙
𝑖,𝑗
∗ 𝑎
𝑗
∑
𝐽
1
𝑎
𝑗
𝑑
𝑖
(
𝑡
)
= 𝑑
0
+ 𝑃
𝑖
× 𝐸
𝑖
− Ω
𝑖
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
44
Network Realisation
True Potential Network True Realised Network
Environmental Preference Similarity
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
45
Network Inference
COOCCUR
HMSC
III
How accurate is network inference?
True Realised Network
Accuracy: "
7/10 70%
Accuracy: "
0/10 0%
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
46
Inference Accuracy
Network Inference Accuracy (n = 975)
COOCCUR
HMSC
True Positive
False Positive
True Negative
False Negative
True Absent
False Absent
III
How accurate is network inference?
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
46
Inference Accuracy
Network Inference Accuracy (n = 975)
COOCCUR
HMSC
1. Network inference accuracy varies widly.
III
How accurate is network inference?
2. Network inference is more accurate when accounting
for environmental preferences.
Tests for the impact of the co-occurrence-performance
spectrum are still outstanding.
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
47
Accuracy Likelihood
Positive
Negative
Absent
IV
What drives differences in
network inference accuracy?
These models are
currently still
compiling.
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
47
Accuracy Likelihood
Positive
Negative
Absent
IV
What drives differences in
network inference accuracy?
3. Correct identification of an association strongly depends on its strength and the differences in
environmental preference of the association partners.
These models are
currently still
compiling.
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
48
Implications macroecological network inference
48
There exists untapped development potential and need regarding ecological network
inference which may be addressed using simulation-based validation approaches.
Ecological network inference ought to be used with care at macroecological scales. To do so, I recommend
alinging method choice with research questions and data availability at the scale of assessment.
Inference of networks using empirical data results in cross-scale
inconsistencies with regards to inferred networks and their topologies.
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
49
IMPLICATIONS & THE BIGGER PICTURE
Paradigm Shifts in Macroecology & Ecological Network Research
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
50
Implications & Development Needs
Updating Macroecological Rersearch Practices.
Chapter I
Using Ecological Networks as Forecast Tools.
Chapter II
Inferring Biological Interactions from Proxies.
Chapter III
Macroecologically must shift from static data products to data streams like
KrigR to address the research questions of the Anthropocene.
I continue to develop of KrigR:
- Widened support of data sets
- Shiny App deployment for easier access
Adopting ecological networks as forecasting tools while accounting for
network resilience mechanisms will enhance biodiversity projections.
NetworkExtinction shortcomings:
- Purely reductionist (no invasion events)
- Rewiring is assumed to be realised fully
and instantly
Inferring Biological Interactions from Proxies.
Avenues forward:
- Network inference method development
- Open Science standards for reporting &
integrating ecological interactions
Ecological network inference approaches ought to be be validated using
synthetic data before application to empiric research questions.
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
51
Everything is a network – Let’s understand them better!
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
52
Supplementary Slides
17/08/2023
Erik Kusch
STADIS Tangled Bank Seminar – Ecological Networks & Macroecological Research
53
https://www.freepik.com/
Icons
https://www.flaticon.com/authors/ddara
https://www.flaticon.com/authors/vectors-
market
https://www.flaticon.com/authors/smalllikeart
https://www.freepik.com/
https://www.flaticon.com/authors/ddara
https://www.freepik.com/
https://www.freepik.com
https://www.flaticon.com/authors/maxim-basinski-premium
https://www.flaticon.com/authors/flat-icons
https://creativemarket.com/Becris
https://www.freepik.com
https://www.freepik.com
https://www.freepik.com/author/starline